Ok so has anyone explained what "AI processor* actually means

Ok so has anyone explained what "AI processor* actually means

Nothing Ever Happens Shirt $21.68

Ape Out Shirt $21.68

Nothing Ever Happens Shirt $21.68

  1. 4 weeks ago
    Anonymous

    It means I can make images of more than just your mom sucking my wiener

  2. 4 weeks ago
    Anonymous

    It has an AI accelerator, akin to the tensor cores on the rtx gpus.

    • 4 weeks ago
      Anonymous

      Is it actually any good

      • 4 weeks ago
        Anonymous

        Only for light AI tasks, are you really expecting 4090 levels of performance from a 65W CPU?
        Still, if you pair it with an nvidia gpu it may be able to do some work on parallel with the nvidia card.

        • 4 weeks ago
          Anonymous

          >Only for light AI tasks
          Like?
          >are you really expecting 4090 levels of performance from a 65W CPU?
          No and I dont know how anything I've said in those five words would give you that idea

          • 4 weeks ago
            Anonymous

            >Like
            Probably some shit like pattern recognition and computer vision, and I'd guess that if this becomes popular some applications like photoshop would use it for AI powered image effects.
            Personally I can't see a single task one of these would be actually useful outside little gimmicks no one will use after a week.

            • 4 weeks ago
              Anonymous

              >Personally I can't see a single task one of these would be actually useful outside little gimmicks no one will use after a week
              Off the top of my head:
              >Voice recognition and TTS
              >OCR
              >Translation
              >Spellcheck and suggestions
              >Search and summarize
              >Tagging and sorting
              >Probably some sort of heuristic antivirus bullshit
              >Drawing stuff, like you said
              There's not really a lot of stuff that people do with their computers that can't be enhanced in some way with AI.

              • 4 weeks ago
                Anonymous

                Ok but processors can do all that fine now

              • 4 weeks ago
                Anonymous

                Extremely simplified: Basically "AI" is a long multi-branched path (neural network) of possible choices and the CPU calculates the best (or most optimal) choice via previously trained data. Think chess engines, which were "peak AI" in the 80s, but more complicated. So an "AI CPU" would have added instructions to make the calculations faster and use less power for more advanced "AI" models.

                Actually tons of shortcuts are used to get it to work on consumer devices, thus diminishing the quality, accuracy and efficiency of the model.
                t. cs student in goymany

      • 4 weeks ago
        Anonymous

        Sure, but it's not for stuff like LLMs or Stable Diffusion. The point is to handle minor AI/ML tasks that will presumably be added to future operating systems, like automatically tagging and sorting media.

      • 4 weeks ago
        Anonymous

        It will probably save MS money by having you preprocess before sending requests off to copilot.

      • 4 weeks ago
        Anonymous

        1 it/s for Stable Diffusion when using the NPU alone. thats as good as a 980 GTX, while using just 10 watt. i'd say that good enough for generating coomshit in the background while doing other things.

  3. 4 weeks ago
    Anonymous

    ai hype word for integer math lol

    • 4 weeks ago
      Anonymous

      Is that it? Not even like matrix math?

      • 4 weeks ago
        Anonymous

        AMX is Intel's extension for that https://www.intel.com/content/www/us/en/products/docs/accelerator-engines/advanced-matrix-extensions/overview.html

  4. 4 weeks ago
    Anonymous

    Maybe AVX4096 or something like that? Or parallel instruction pipelines instead of full multicores to stream data through compute graphs asynchronously. I don't know I'm just speculating.

    What do the Tensorflow and Pytorch teams need for good hardware acceleration?

  5. 4 weeks ago
    Anonymous

    It has what the Snapdragon and M1 CPUs had from the beginning

    It's just intel catching up, slow like a dinosaur...

  6. 4 weeks ago
    Anonymous

    lots of alu's

  7. 4 weeks ago
    Anonymous

    It's a GPU with instructions related to graphics stripped
    t. works with NPUs

Your email address will not be published. Required fields are marked *